More than 70% of US grid interconnection requests are ultimately withdrawn due to capacity constraints, according to Berkeley Lab data.1 That bottleneck now sits at the center of a $5.2 trillion AI data center investment cycle — and threatens to compress returns for the hyperscalers funding it.2
Goldman Sachs projects global data center power demand will surge 165% by 2030 versus 2023 levels.3 The US power grid was designed for 1-2% annual demand growth. The gap between those two trajectories is where hyperscaler capex plans collide with physical reality.
Microsoft, Amazon, and Alphabet have each signaled multi-hundred-billion-dollar infrastructure commitments in recent quarters. But committed capex and deployed capacity are not the same thing. When interconnection queues stall or collapse — which Berkeley Lab data shows they do at a 70%-plus rate — announced projects become stranded capital.1
Investor Kevin O'Leary has argued that 50% of planned US data centers will never be built due to grid constraints.4 Even at lower failure rates, the arithmetic affects financial models. A 30-50% curtailment of the $5.2 trillion buildout redirects capital without producing proportional revenue-generating assets.2
The financial risk compounds across the supply chain. Hyperscalers booking land, signing power purchase agreements, and ordering hardware before interconnection approval creates cost exposure that precedes any revenue. If projects cancel post-commitment, write-downs follow.
Investors monitoring hyperscaler earnings should track divergence between capex guidance and actual data center commissioning rates. FERC interconnection queue approval rates are a leading indicator. When withdrawal rates hold above 70%, the probability of announced buildout milestones slipping increases materially.1
The US grid's structural limitations are not temporary. Upgrading transmission infrastructure takes 5-10 years under normal regulatory timelines. AI demand is accelerating now. That mismatch creates a sustained headwind for capex-to-capacity conversion rates through at least 2030.
Hyperscalers with diversified geographic exposure — including European and Asian grid markets — carry less concentration risk. Those weighted heavily toward US buildout face the highest exposure to interconnection bottlenecks.1
The $5.2 trillion figure represents announced intent, not guaranteed deployment.2 Grid physics may prove to be the binding constraint that financial models have not yet fully priced.
Sources:
1 Berkeley Lab, US Grid Interconnection Queue Analysis
2 Industry AI Data Center Capex Forecasts, 2026
3 Goldman Sachs, Global Data Center Power Demand Forecast, 2026
4 Kevin O'Leary, Public Commentary on US Data Center Grid Constraints
